Amino acid repeats (AARs) are segments of proteins made up of simple patterns of amino acids, often strings of a single amino acid. They have long been recognized to be common features of eukaryotic proteins 1234. Polyglutamine repeats, the most intensively studied class because of their association with human diseases such as Huntington's 5, tend to be evolutionarily labile, especially when encoded by pure repeats of the codon CAG 67. Because of this lability, AARs have often been considered to be evolutionarily neutral structures 8. However, a number of experimental studies 9101112 suggest that AARs play an important role in protein function. Studies of the functions of AAR-containing proteins also suggest that they are preferentially found within certain classes of proteins. From the earliest reports through to the most recent genome-wide surveys in Saccharomyces cerevisiae 31314 and mammals 15 a consistent pattern of association with transcription has emerged for the most common tandem repeat types. Additional associations, notably with protein kinases 13, suggest possible involvement in cellular signaling networks, which in turn suggest that repeats could play a significant role in the evolution of such networks 16. Finally, studies of the relationship between morphology and repeat length in dog breeds 17 have shown that variation at repeat loci can have evolutionarily significant effects on phenotype. Polyalanine repeats have also been found to be involved in a number of genetic diseases, in this case involving developmental defects 18. Removing a polyalanine tract from murine Hoxd-13 has a direct effect on bone phenotype 19, again indicating involvement of an AAR in an important biological process.

AAR size difference between orthologous human and mouse proteins correlates with protein nonsynonymous substitution rate 20. A study of the factors contributing to the evolutionary expansion of polyglutamine repeats in a limited number of human-mouse orthologues 21 concluded that labile repeats, which are encoded by homogeneous runs of a single codon 6, have a strong tendency to arise in regions of proteins subject to weaker purifying selection than the protein as a whole, while repeats that are more conserved did not show this tendency. This has been supported recently by a large-scale study of human, mouse and rat repeats 22. These observations suggest a model for repeat evolution whereby initially labile repeats become fixed when they reach some optimal length range 21. Human polyglutamine disease genes might then be still evolving towards such an optimum.

Intrinsically unstructured regions (IURs), also called disordered regions, are regions of protein, ranging in size from short loops to complete proteins, that do not form a compact tertiary structure under normal solvation conditions 23. They have been suggested to be involved in protein-ligand binding, including protein-protein interactions, forming compact structures only when bound to a cognate ligand 24. Tompa 25 pointed out that many IURs contain AARs and suggested that IURs may evolve to a considerable extent by the expansion of such repeats. Disordered proteins - that is, proteins primarily made up of IURs - have also been suggested to have lower sequence complexity than ordered proteins 26. Tompa's suggestion 25 would be consistent with the relatively rapid sequence evolution of many IURs 2728, the observation that highly connected (hub) proteins in protein interaction networks appear to be enriched in AARs and in proteins containing IURs 29, and the suggestion that evolution of AARs could have an effect on network evolution by altering protein-protein affinities 16. As Tompa 25 analyzed only a relatively small set of IURs, his hypothesis raises the question whether AARs show a preferential location in IURs, and whether any such preference could account for the evolutionary properties of the bulk of AARs in a proteome. Such a preference would be consistent with hypotheses on the causation of triplet expansion diseases that invoke destabilization of protein structure as an important causative factor 18.

A variety of computational methods exist to detect repeated sequences in proteins. These range from SEG, which looks for regions of low complexity 30, to alignment-based approaches 31. Here we use an extended definition of amino acid repetition that includes cryptic repeats as measured by the program SIMPLE, which we have previously used to look at AARs in the yeast proteome 32, as well as tandem AARs. This allows us to study repeats below the normal threshold taken for tandem repeats (five amino acids) and regions with significant biases in amino acid content that are not tandem in nature but may have originated from tandem repeats (C4 repeats; see Materials and methods for more detail).

Using a set of orthologues to human genes from four species (chimpanzee, mouse, rat and chicken; Pan troglodytes, Mus musculus, Rattus norvegicus and Gallus gallus) we show that the most common AARs show strong preferences to be located within IURs in all five proteomes. We also confirm that sequences flanking AARs evolve more rapidly than the remainder of their respective proteins. We conclude that the forces shaping the evolution of IURs and AARs are strongly linked, although AARs are present in only a subset of IURs.

Although tandem repeats of amino acids are easily recognized features of proteins and have been extensively studied, protein sequences show more widespread repetitive features. This is shown by the high proportion of proteins containing repetitive segments - approximately 50% as measured by SEG 30 and over 70% of the S. cerevisiae proteome as measured by SIMPLE 32. In this study we have compared the frequencies of tandem repeats with those of C4 repeats (repetitive regions with a local overrepresentation of motifs of length four residues) using SIMPLE, which has the advantage that it identifies explicitly the overrepresented motif in a given region. We have carried out this comparison in a large set of proteins orthologous between four mammals and chicken, which is the most closely related non-mammalian species with a sequenced genome. This allows us to compare repeat frequencies both between types and between species.

After excluding C4 motifs that overlap tandem repeats, many of the C4 motifs detected in these genomes are clearly related to common tandemly repeated amino acids (six of the seven most common tandem amino acid types in Figure 1a are mirrored by the six most common homogeneous C4 repeat types in Figure 1b), suggesting that the underlying mechanisms that gives rise to them is similar. This is also reflected in the high correlations seen between the frequencies of tandem repeats and their respective homogeneous C4 repeats. Tandem AARs most likely evolve by replication slippage, as they evolve more rapidly if they are encoded by pure codon repeats than interrupted codon repeats 613. Dieringer and Schlötterer 46 introduced a novel, slippage-related process they called indel slippage that acts in a non-repeat-length-dependent manner on repeated motifs as short as a single nucleotide. Such a mechanism could contribute to the evolution of C4 repeats and other cryptically repetitive sequences 4748 and could give rise to differences in the frequencies of tandem and cryptic repeats.

The biggest difference in frequency between tandem and cryptic repeats was seen for Leu, which is rare among C4 repeats. In addition, Q₄ repeats are by far the most common class of C4 repeats while Gln is only the seventh most numerous class of tandem repeat in our sample. These large differences could reflect differences in underlying mechanisms (although this seems superficially unlikely as Q tandem repeats are known to undergo rapid evolution 649) but could also reflect differential selective forces (acting strongly against L₄ repeats and Q tandem repeats but less so against their counterparts).

Repeat frequencies were highly similar between the mammals, but the chicken proteome showed a distinct frequency distribution in which most repeat frequencies were lower. A partial exception to this pattern were tandem S repeats, which although rarer in chicken than in mammals, were the most common class in the chicken proteome. A trivial explanation for these differences could be the currently lower quality of the chicken genome sequence. However, this is unlikely to be the main explanation as the dataset we used contained only clearly identifiable orthologues. Another, and more interesting, possibility is that the lower frequency in chicken is the result of the general reduction of genome size in birds. The chicken genome is approximately one-third the size of the human genome 50 while bird genomes in general are approximately half the size of mammalian genomes 51. Analysis of the evolution of bird genome size indicates that genome shrinkage took place in the saurischian lineage leading to the birds circa 200 to 300 million years ago and that this was accompanied by a reduction in the genome fraction of repetitive elements 51. A global correlation of genome sequence repetition with genome size has also been described 5253. The lower frequency of amino acid repeats in chicken proteins may therefore reflect a parallel process of loss of transposable elements and tandem and cryptic repeats in that evolutionary lineage. A possible explanation for the stronger conservation of S repeats between mammals and chicken than other repeat types is that they play a less dispensable role in protein function; serine-rich domains (RS domains) are intimately involved in alternative splicing 54 and it is possible that this role is sufficiently important to ensure their retention.

Previous analyses of the evolution of Gln repeats have suggested that in the early stages of their emergence, when encoded by pure codon repeats, they appear preferentially in regions of proteins that are subject to relatively low levels of purifying selection (that is, regions that evolve more quickly than the rest of the protein) 72122. In this study we have analyzed the evolution of regions flanking tandem and C4 AARs in human-rodent and human-chicken comparisons and show the same trend, confirming that the majority of tandem and C4 repeats in proteins emerge in rapidly evolving subregions. We also confirm earlier suggestions 2122 that conserved repeats lie in relatively more conserved protein subregions than non-conserved repeats and show that conserved AARs tend to lie in more conserved proteins than non-conserved AARs. In addition, we observe elevated sequence differences around conserved repeats of both types, although this elevation is less extreme than is observed for non-conserved repeats. The latter result differs from a previous study that did not find a difference between flanking regions and the remainder of the proteins for conserved AARs 21. However, that study only considered a relatively small number of proteins and so most likely failed to detect this difference due to a lack of statistical power. Generally, the results are consistent with a model of repeat evolution whereby repeats tend to emerge in less-conserved regions of proteins and become frozen in length as they reach a length at which they are close to a threshold at which they may cause deleterious phenotypes 1621 but they also suggest that the regions in which repeats become fixed may continue to evolve relatively rapidly after repeat fixation.

IURs are regions of proteins that do not form stable tertiary structures under native conditions. Analyses of the extent of disorder in whole genomes suggest that in eukaryotes more than 40% of proteins are either completely disordered or contain significant regions of disorder 4041. In this dataset we find 34% of proteins to contain IURs of length > 50 and 85% to contain an IUR of length > 10. These regions are thought to form flexible regions of proteins that might have a number of functions, including binding to other proteins and small molecules and providing flexibility in multidomain proteins. In an analysis of repeat content of a relatively small number of intrinsically unstructured protein regions, Tompa 25 identified an apparently strong role for AARs in IUR evolution. The definition of 'repeats' in his analysis is different from ours as it included longer, complex repeated motifs as well as simple sequence repeats, but some simple sequence repeats did appear in his results. This raises the question whether there is a real association of simple AARs with IURs, and whether an association of this type can account for the evolutionary dynamics of AARs. Here we have investigated this by considering the overlap between tandem and C4 repeats and, first, domains identifiable searching the SUPERFAMILY and InterPro databases, and second, unstructured regions predicted by the RONN predictor. The majority of AARs, with the exception of L tandem repeats, lie within IURs predicted by RONN (Figure 6).

We obtained inconsistent predictions on the level of structure shown by A repeats. They were predicted to be predominantly unstructured by two methods, RONN and DISOPRED, but not by a third, IUPRED. This disagreement may reflect the different methodologies employed by the different algorithms as IUPRED takes account of the chemical characteristics of the sequence being analyzed whereas RONN and DISOPRED use structural analyses of proteins. The ambiguous position of A in these analyses is interesting in the light of its role as the second major cause of human repeat expansion disease, after Q. Gln repeats are notable in showing markedly higher proportions of disorder as tandem repeats than as C4 repeats, suggesting that expansion of Q repeats could have a destabilizing effect on proteins, as suggested previously 18.

Seven of the eight most common tandemly repeated amino acids in our dataset correspond to the seven disorder-promoting amino acids defined by Dunker et al. 55. Lise and Jones 56 in their study of common amino acid patterns in unstructured regions also identified a number of patterns similar to the most common C4 repeats, notably E- and P-rich regions. A strong element of the purifying selection acting against the emergence of AARs within folded regions of proteins therefore appears to be selection against their propensity to lower the stability of these regions. Interestingly, as noted by Kreil and Kreil 57, N repeats are much rarer than Q repeats - indeed, in our analysis of human proteins we found only four tandem N repeats. This observation may reflect the propensity of Asn to promote order 55 and consequent purifying selection acting against the appearance of N repeats in unstructured regions. A similar argument may apply to D and E repeats - Glu, which is common in AARs, is disorder-promoting whereas Asp, which is rare in AARs, is not. In this context, it is noteworthy that although E repeats are the most common class in mammals and the most often predicted to be unstructured, they are also, after L repeats, the class most commonly found associated with SUPERFAMILY and InterPro domains. This raises the question whether the domains in which they are located tend to be close to the threshold of instability. Mean RONN scores of domains containing E repeats are 0.44 for SUPERFAMILY and 0.46 for InterPro domains. These compare to means for all domains containing repeats of 0.43 for SUPERFAMILY domains and 0.41 for InterPro domains. The mean for E repeats in SUPERFAMILY domains is typical of all repeat-containing domains, but that for InterPro domains is the highest amongst all repeat types. As most of the domains containing E repeats are InterPro and not SUPERFAMILY domains, this raises the possibility that some E repeat-containing InterPro domains are relatively unstable.

L tandem repeats form interesting exceptions to the general association of AARs with unstructured regions as they are predicted to be 100% structured. The amino acids found in tandem repeats tend to be hydrophilic; all the most hydrophilic amino acids 58 are found in the class of common tandem AARs - the only strongly hydrophobic amino acid in this class is Leu. Hydrophobic amino acids tend to occupy buried positions within proteins, so it is not surprising that Leu repeats show a high propensity to be structured. In earlier analyses, Leu repeats have been found to be concentrated close to the amino termini of proteins 1559, presumably forming part of the hydrophobic region of signal sequences, although Leu may also contribute to transmembrane segments of proteins and more generally to protein cores and stabilizing secondary and tertiary structure 59.

The majority of AARs have arisen during evolution within protein regions with the characteristics of IURs. This is true both of tandem and cryptic repeats, which have many common characteristics such as relative frequency and, to a lesser extent, GO associations. The dynamics of the evolution of most AARs are, therefore, likely to mirror those of IURs. Some, but not all, IURs, evolve more rapidly than the proteins they are part of 2728. Despite this, our results suggest that only a small subset (no more than 15%) of IURs contain AARs. This raises the question whether there are specific subclasses of rapidly evolving IURs that have a higher propensity to evolve AARs. As AARs tend to be associated with transcription and cell signaling, it is possible that proteins with these types of functions have particular types of IUR that might predispose them to evolve repeats.

IURs are thought to play an important role in protein-protein interactions. Repeat accumulation may, therefore, play a role in the evolution of protein-protein interactions in transcriptional and signaling networks by expanding the repertoire of disordered regions. Because they evolve rapidly, repeat sequences potentially provide a means for organisms to rapidly tune their transcriptional and signaling protein-protein interaction networks 16.

Leu (and Ala) repeats form a special class in being hydrophobic amino acids that commonly form repeat structures. Leu repeats are consistently predicted to be structured, and Ala repeats often are. Glu repeats, which are very common, are also often found within structured regions, although Glu is disorder-promoting. Further studies of the evolution of these repeat classes are therefore merited as repeat variation in structured regions may be expected to have significant effects on protein structure and/or stability.

AAR: amino acid repeat; GO: Gene Ontology; IUR: intrinsically unstructured region; RONN: Regional Order Neural Network.

MS carried out most of the data acquisition and analysis and drafted parts of the manuscript. JMH supervised the project, carried out some of the analysis, and compiled the final manuscript.